Attribution as Retrieval: Model-Agnostic AI-Generated Image Attribution

Authors: Hongsong Wang, Renxi Cheng, Chaolei Han, Jie Gui

Published: 2026-03-11 09:40:00+00:00

Comment: To appear in CVPR 2026, Code is at https://github.com/hongsong-wang/LIDA

AI Summary

This paper presents LIDA (Low-bIt-plane-based Deepfake Attribution), a model-agnostic framework that redefines AI-generated image attribution as an instance retrieval problem rather than classification. LIDA generates low-bit fingerprints from images and trains an attribution encoder through unsupervised pre-training and subsequent few-shot adaptation. The approach achieves state-of-the-art performance for both deepfake detection and image attribution under zero- and few-shot settings.

Abstract

With the rapid advancement of AIGC technologies, image forensics will encounter unprecedented challenges. Traditional methods are incapable of dealing with increasingly realistic images generated by rapidly evolving image generation techniques. To facilitate the identification of AI-generated images and the attribution of their source models, generative image watermarking and AI-generated image attribution have emerged as key research focuses in recent years. However, existing methods are model-dependent, requiring access to the generative models and lacking generality and scalability to new and unseen generators. To address these limitations, this work presents a new paradigm for AI-generated image attribution by formulating it as an instance retrieval problem instead of a conventional image classification problem. We propose an efficient model-agnostic framework, called Low-bIt-plane-based Deepfake Attribution (LIDA). The input to LIDA is produced by Low-Bit Fingerprint Generation module, while the training involves Unsupervised Pre-Training followed by subsequent Few-Shot Attribution Adaptation. Comprehensive experiments demonstrate that LIDA achieves state-of-the-art performance for both Deepfake detection and image attribution under zero- and few-shot settings. The code is at https://github.com/hongsong-wang/LIDA

Key findings

LIDA achieves state-of-the-art performance for both AI-generated image detection and attribution across zero-shot and few-shot settings on the GenImage and WildFake datasets. The low-bit generative fingerprints proved highly effective in capturing model-specific noise patterns, leading to significant improvements over baseline methods. The framework also demonstrates strong robustness to image degradations like Gaussian blur and JPEG compression, while maintaining high practical efficiency.

Approach

The authors formulate AI-generated image attribution as an instance retrieval problem, enabling model-agnosticism. Their LIDA framework first extracts 'low-bit generative fingerprints' from input images. An adapted ResNet-50 serves as the attribution encoder, trained via unsupervised pre-training on real images, followed by few-shot adaptation using a combination of center loss for attribution and a real-prototype-based contrastive loss for deepfake detection.

Datasets

GenImage, WildFake, ImageNet

Model(s)

ResNet-50 (adapted)

Author countries

China

← Previous